Post AcjxrMnEa3MTLPyEd6 by aaronesilvers@jawns.club
(DIR) More posts by aaronesilvers@jawns.club
(DIR) Post #AcjxrKnW02Oj9i5Qn2 by saila@mastodon.social
2023-12-11T17:42:00Z
0 likes, 0 repeats
I would love to read @clive's take on #StevenJohnson work on #Google’s #NotebookLM Both are part of the few non-fiction writers who have continued to be in my must read list (as in, whatever they published I trust enough to spend my time reading and learning from them) because in part they both offer smart views that on the process of thinking about complex ideas https://www.wired.com/story/googles-notebooklm-ai-ultimate-writing-assistant/#AI #LLM
(DIR) Post #AcjxrLoyC79iKVm6hE by aaronesilvers@jawns.club
2023-12-12T17:04:34Z
0 likes, 0 repeats
@saila @clive so... i'm 24 hours into #NotebookLM. It's intriguing to me as a skeptic of the OpenAI faucet-full-on approach. As a longtime fan of Steven Johnson, it's kinda exactly what I'd hope it'd be, and right now nothing much more. It generates from the corpus you feed it. It cites how/where it's coming up with the generated content. It allows you to query the corpus directly. It can be useful as a tool to help me upcycle learning curricula into new, more engaging media
(DIR) Post #AcjxrMnEa3MTLPyEd6 by aaronesilvers@jawns.club
2023-12-12T17:06:17Z
0 likes, 0 repeats
@saila @clive AND... that Steven is on the discord actively that supports the tool, it kinda inspires some confidence as a right-sized project/slice of AI with enough well-intentioned stakeholders present with users to maybe... maybe... it both does not suck AND could be good?
(DIR) Post #AcjxrNVttz5fZwruFc by saila@mastodon.social
2023-12-12T17:14:01Z
0 likes, 0 repeats
@aaronesilvers That's so good to hear — I feel like it's kinda of a perfect playground for him to be working in and will interested to try it myself if/when it's available more widely
(DIR) Post #AcjxrOGh60WLv4lHBg by clive@saturation.social
2023-12-12T17:34:21Z
0 likes, 0 repeats
@saila @aaronesilvers I haven't tried it yet, so I don't know what it's like!It didn't surprise me to hear that Steven is working with that project -- it really is something he's been pondering for *decades* nowWith LLMs, I myself have been most intrigued by their potential for autosummarization and connection-making with retrievel-based generative techniquesIt seems like there's enormous potential here for helping us work with huge corpuses of research ... 1/x
(DIR) Post #AcjyEwrMeTMNrb9B2m by clive@saturation.social
2023-12-12T17:38:38Z
0 likes, 0 repeats
@saila @aaronesilvers I wanna give NotebookLM a shot!My only reservation is that I'd be nervous putting transcripts of interviews into itMy interviews, like those of most journalists, contain moments of mundane, uncontroversial on-the-record comments that exist cheek-by-jowl with stuff that's off the record, and which I thus would never circulate publicly, everSo I would be *super* hesitant to have a system crawling all over those 2/x
(DIR) Post #AcjyQ8ynvjfqddxuzY by clive@saturation.social
2023-12-12T17:40:42Z
0 likes, 0 repeats
@saila @aaronesilvers Whatever Google promises it isn't doing with that data now (i.e. using it to train new LLMs) could change in the future... and it would be a complete betrayal of my promises of confidentiality and off-the-record privacy to my interviewees should Google decide, four years from now, daaaaaaaaamn we need fresh text to train new models, let's change the terms of service for NotebookLM under *cover* of *darkness* ...... and boom, that stuff is in a training corpus3/x
(DIR) Post #Acjyc2RK4VOlpVybGS by clive@saturation.social
2023-12-12T17:42:51Z
0 likes, 0 repeats
@saila @aaronesilvers Since LLMs have shown a penchant for regurgitating private information word-for-word, given the right prompts -- witness OpenAI's recent scramble to respond to the "repeat this word forever" hack ...... I basically figure I *wouldn't* be able to use NotebookLM to process any of my research materials that include interview ...... which is, for me, kinda useless: I did about 200 interviews for my last book, which ranged anywhere from 3K to 15K in transcript length4/x
(DIR) Post #AcjyoKo1c2KEGaSURU by clive@saturation.social
2023-12-12T17:45:03Z
0 likes, 0 repeats
@saila @aaronesilvers Being able to use an LLM to help autosummarize them, find interesting connections in the material, etc: That's gold!But I wouldn't ever want to have that stuff in a cloud environment where a tech giant could abruptly decide to phagocytose it for trainingSo if someone did a locally-run verision of this, I'd be all over it, in a heartbeatI've been playing with llamafile lately, which is just such a local-only LLM But alas, it's pretty bad at summarization5/x
(DIR) Post #Acjz0Da0Iy1RKWZCPQ by clive@saturation.social
2023-12-12T17:47:11Z
0 likes, 0 repeats
@saila @aaronesilvers Anyway, that's my 2 cents on NotebookLMI think the idea has great promise, and Steven's involvement is undoubtedly quite good -- he deeply understands the type of product he himself wants to exist!I'll likely give it a shot with low-stakes research that's all open-source -- open-source in the "intelligence community" meaning of the word, i.e. containing info that's inherently publicly available: Book and article snippets, my notes thereon6/x
(DIR) Post #Acjz7iLCKlTQlGDROC by clive@saturation.social
2023-12-12T17:47:30Z
0 likes, 0 repeats
@saila @aaronesilvers But I don't think I could trust it for the real meat of my reporting -- the interviews7/7
(DIR) Post #Acjz7jEqzpzdXsFt8i by clive@saturation.social
2023-12-12T17:47:57Z
0 likes, 0 repeats
@saila @aaronesilvers (And thank you for the kind words, Craig!!)
(DIR) Post #AcjzmmKsO00N1Xe8MS by saila@mastodon.social
2023-12-12T17:52:37Z
0 likes, 0 repeats
@clive Thanks — appreciate the insights and thoughts! And agree, having him involved so closely with it suggests that some of the more unique needs of a research+serendipity tool might unveil some interesting new features other AI tools could adopt, too /cc @aaronesilvers
(DIR) Post #Ack0mxPRJZEFK2QiBM by saila@mastodon.social
2023-12-12T17:49:01Z
0 likes, 0 repeats
@clive Yes to this — the confidentiality thing is a huge issue, especially, as you mention, it's governed by a "trust us" covenant that seems to be regular broken at willWould love to use a locally-run, maybe even E2E LLM that's optimized for transcriptions/summarizations and the kind of "did you also know…" connections that NotebookLM seems to be getting built for /cc @aaronesilvers
(DIR) Post #Ack0myLvo616FRnQLw by clive@saturation.social
2023-12-12T17:56:46Z
0 likes, 0 repeats
@saila @aaronesilvers The fact that llamafile works so well on laptops -- and grant you, I'm using a brand-new M2 Mac, so it'd be slower on nearly any other machine -- makes me think someone really could make a local-only version of NotebookLMThe only challenge is ...... so few startups *want* to make locally-run toolsThe biz models have so enSAASified that they have all cloudifiedMaybe I'll have to take a few months off work and build it myself 😆
(DIR) Post #Ack19JKOVCjJfkjFk8 by saila@mastodon.social
2023-12-12T18:11:09Z
0 likes, 0 repeats
@clive I’d Kickstart that! (Also now tempted to try llamafile, too)/cc @aaronesilvers
(DIR) Post #Ack1HNObGJBYbaXfiS by clive@saturation.social
2023-12-12T18:12:37Z
0 likes, 0 repeats
@saila @aaronesilvers yeah, give it a try!It was really quite easy to set up, just a handful of terminal commands and presto
(DIR) Post #AckCzlYhsusBmbdf28 by aaronesilvers@jawns.club
2023-12-12T19:09:38Z
0 likes, 0 repeats
@saila @clive would 100% fund that kickstarter. having this kind of tool, but on and constrainable to what's on the hard drive itself, is the dream
(DIR) Post #AckCzmPAjqqAPKBYoK by clive@saturation.social
2023-12-12T20:22:42Z
0 likes, 0 repeats
@aaronesilvers @saila yep yep
(DIR) Post #AcmljJZc3FV14AsC6i by clive@saturation.social
2023-12-14T02:01:20Z
0 likes, 0 repeats
@briansholis @saila That’s a great writeup!